AAAAARRRGGGHHH.... Still confused...

MuFu

MuFu

Moderator
#21
Good work RDX, that's interesting reading.

MuFu.
 
RDX

RDX

Proud H-Bomb owner
#23
I wonder if that's a typo. That card in Tom's benchmark is clocked at 380/675.

All sapphire cards that I've seen are clocked at 325/500, 325/540 , 325/580, or 380/600 and have 128-bit memory with the last one being available only with 2 analog outputs (vs. 1 analog, 1 DVI).

Powercolor makes a card that is clocked at 380/600 (but has a DVI out unlike the sapphire card clocked at this speed.), and has 128 bit memory. Their other card is clocked at 380/680 and has 256-bit memory. This card is sometimes called the 9800se gold.

I have never seen a sapphire card clocked at 380/675 or a any other 128-bit 9800se clocked this high. The only cards produced with those memory speeds are the 256-bit powercolor 9800se gold, and strictly for our European friends, the Hercules 9800se AIW.
 
O

ol' man

workin fingas to the bone
#24
Originally posted by RDX
I wonder if that's a typo. That card in Tom's benchmark is clocked at 380/675.

All sapphire cards that I've seen are clocked at 325/500, 325/540 , 325/580, or 380/600 and have 128-bit memory with the last one being available only with 2 analog outputs (vs. 1 analog, 1 DVI).

Powercolor makes a card that is clocked at 380/600 (but has a DVI out unlike the sapphire card clocked at this speed.), and has 128 bit memory. Their other card is clocked at 380/680 and has 256-bit memory. This card is sometimes called the 9800se gold.

I have never seen a sapphire card clocked at 380/675 or a any other 128-bit 9800se clocked this high. The only cards produced with those memory speeds are the 256-bit powercolor 9800se gold, and strictly for our European friends, the Hercules 9800se AIW.
Well what the hey. I wish someone could do a proper benchmark review without bullshitting around. I have no idea why it says it is running at that speed.

Well no matter what 380MHz for a r350 core does not even make it break a sweat. It would probably do another 50~100MHz especially with only 4 pipes. If it really was clocked that high then that is the reason. And I have heard they put pretty decent memory on them too at times but also the crappy tsop.
 
RDX

RDX

Proud H-Bomb owner
#25
Well what the hey. I wish someone could do a proper benchmark review without bullshitting around.
I would second that. It's even harder trying to figure out what vendors are selling when they advertise a 9800se. It's seems like every one I look at has a description for a sapphire 9800se, a product code for a 9800se gold, a picture of a sapphire 9800se @ 300/500 and 256-bit memory claims in the specs. :rolleyes:
 
O

ol' man

workin fingas to the bone
#26
Thats why I would go with the egg. $152 is a very low ball amount. If you pay $170+ you would expect to get a 256bit and if you didn' you would be pissed. You may get a tsop:( The egg wouldn't BS you around with what they have and if they made a mistake they would take care of it the right way I think.
 
N

natiels

New Member
#27
This is an interesting discussion. I just purchased a refurbed powercolor 9800SE from newegg.com for $132. I figured it was probably just sent back because it wasnt a 256-bit version. When i got it i found out that it wasnt. It was only the 128-bit but it modded perfectly.

I never even installed the normal drivers. I went straight to the omega drivers that enable all 8 pipelines. at stock speeds of 380/300 i got 13500 on 3dmark01. I was a bit disappointed since i had read like many of you about overclocked 9500pro's getting really nice scores. i even saw one article where a 9500 pro that was overclocked was doing about as well as a 9700 at stock speeds. Previously I had a 9500np with a 256-bit bus and 4 pipelines clocked at 375/290 and it scored 13000.

I will install the 4 pipeline drivers when i get home and see what kind of results i get and post em here to show you the difference between the powercolor 128-bit 9800SE at 4 pipelines.

I wonder if there is something holding this card back. perhaps something in the drivers or the bios. I'll post my results as soon as i get them.
 
RDX

RDX

Proud H-Bomb owner
#28
:) Excellent, it would be nice to get some first hand experience on the matter. BTW, can you tell us the type and rating of the ram on the powercolor card?

And welcome to HardwareCentral!
 
N

natiels

New Member
#30
ok, i ran home for lunch and did some benchmarking.

I have a powercolor 9800SE with 128-bit memory bus and samsung 3.3ns ram. The default clocks are 380 core, 300 mem.

with default clocks i got the following 3dmark 2001 scores
4 pipeline - 12,042 omega drivers
8 pipeline - 13,658 omega softmod drivers
8 pipeline clocked @ 365/371 - 14,218 omega softmod drivers

I have an ASUS A7N8X deluxe 2.0 with FSB at 203mhz and AMD cpu @ 1930mhz. 512megs single channel.

I had to underclock the core in order to overclock the memory at all. It wouldnt break 310 without artifacts otherwise. Once I brought the core down a bit that memory really soared. I am considering a volt mod to get a bit more from the core/mem. Right now I have an old amd HS/F on the core and it keeps it pretty cool compared to the stock hs/f.

The powercolor is definitely the best of the 9800SEs right now with its default clock of 380/300 and the good quality mem they put on the boards. The only problem is that it normally costs like $170-$180 due to the fact that everyone was buying them trying to get the 256-bit models. I got mine as a newegg refurb card for $132. If not for that good price one would be better off buying a 9700 or 9700 pro. I have seen some sites lately having pretty good sales on these cards. A while back gameve.com had 9700 for $170-free shipping. I think they still have Pros left that they are selling for $199.
 
RDX

RDX

Proud H-Bomb owner
#31
That seems like a pretty decent improvement for 3dmark 2001. In the reviews I've seen with that benchmark, there isn't a lot of difference between a 9500np and a 9500pro. Therefore, unlocking your card shouldn't see too much of a difference in that test either.

Do you have any other benchmarks readily available? Something like unreal tournament or aquamark might show the difference between the 4 pipes and the 8 pipes better.
 
O

ol' man

workin fingas to the bone
#32
Originally posted by RDX
That seems like a pretty decent improvement for 3dmark 2001. In the reviews I've seen with that benchmark, there isn't a lot of difference between a 9500np and a 9500pro. Therefore, unlocking your card shouldn't see too much of a difference in that test either.

Do you have any other benchmarks readily available? Something like unreal tournament or aquamark might show the difference between the 4 pipes and the 8 pipes better.
My scores went up 2000 points in 3dmark with the softmod.

Here is a default 9500 vs 9500 softmod.

both cards are clocked at 275/270


Here is my 9500 softmod OC'd vs 9800 pro underclocked. ALl clocks are the same. I had to overclock my 9500 softmod cause my 9800 would get artifacts if I clocked it lower than around 300MHz.

clocks for both cards are at 390/310.


One must keep in mind that a r350 core has a higher IPC than the r300. A 9700pro and a 9800np clock at the same speeds the 9800np will them all as far as I have seen by around 1~3% or better.

I have been busy tonight. Here is what I have done to my radeons.





 
O

ol' man

workin fingas to the bone
#33
Enabling the other 4 pipes made the score go up 2000 points in 3dmark but enabiling the 256 bit bus would make it only go up 1000 points.

It is better have 8 pipes than the extra memory bandwidth.
 
RDX

RDX

Proud H-Bomb owner
#34
Nice cooling there ol' man! So what do you think of unlocking the 9800se? Natiels 9800se with 8 pipes gave him a 13.4% performance increase over 4 pipes. You got a 18% performance boost by softmodding your 9500. Considering other factors, such as him being moderately more CPU limited than yourself, these performance increases are comparable. It looks to me like the only thing really holding the unlocked 9800se back is its lack of memory bandwidth. If you look at the 9500 vs the 9800se the, the core speed difference is over 100mhz; the memory difference is only 30mhz though.

The 256 bit 9800se sees such a big performance jump going from 4 pipes to 8 is because it has tons of memory bandwidth (680 ddr 256-bit). The core is the limiting component on the video card. Dramatically increasing the pixel throughput by doubling the pixel pipes gives you a much larger percentage performance boost compared to doubling the pipelines on a memory limited card.

As you and Natiels demonstrated, moving from 4 pixel pipelines to 8 is really nice; but at those core speeds, the card is starting to become bandwidth restricted.
 
O

ol' man

workin fingas to the bone
#35
Originally posted by RDX
Nice cooling there ol' man! So what do you think of unlocking the 9800se? Natiels 9800se with 8 pipes gave him a 13.4% performance increase over 4 pipes. You got a 18% performance boost by softmodding your 9500. Considering other factors, such as him being moderately more CPU limited than yourself, these performance increases are comparable. It looks to me like the only thing really holding the unlocked 9800se back is its lack of memory bandwidth. If you look at the 9500 vs the 9800se the, the core speed difference is over 100mhz; the memory difference is only 30mhz though.

The 256 bit 9800se sees such a big performance jump going from 4 pipes to 8 is because it has tons of memory bandwidth (680 ddr 256-bit). The core is the limiting component on the video card. Dramatically increasing the pixel throughput by doubling the pixel pipes gives you a much larger percentage performance boost compared to doubling the pipelines on a memory limited card.

As you and Natiels demonstrated, moving from 4 pixel pipelines to 8 is really nice; but at those core speeds, the card is starting to become bandwidth restricted.
Yeah but the memory will still not bottle neck the card. Here is my new rad9500pro @ 410/300. Notice what the xtra 20MHz core speed does for the score vs. the 390/310 rad9500 softmodded. I have a couple of my cas timings set tighter too right now which may account for some of the increase.

 
N

natiels

New Member
#36
Woah, was that bench done on the same PC as the other ones? with just that 20mhz core increase and a drop of 10mhz on the mem you went up almost 3000 marks. That is almost 2000 faster than the 9800 pro was at 390/310. You should try that again with your cas timings set back to what they were for the original tests and see what the score is. I wonder what would happen if you put the bios from that 9500pro into the older softmodded card. Maybe it would make a big jump. If that is the case then we know the bios is holding things back.
 
O

ol' man

workin fingas to the bone
#37
Originally posted by natiels
Woah, was that bench done on the same PC as the other ones? with just that 20mhz core increase and a drop of 10mhz on the mem you went up almost 3000 marks. That is almost 2000 faster than the 9800 pro was at 390/310. You should try that again with your cas timings set back to what they were for the original tests and see what the score is. I wonder what would happen if you put the bios from that 9500pro into the older softmodded card. Maybe it would make a big jump. If that is the case then we know the bios is holding things back.
Sorry forgot to add that the 9800 vs 9500 softmod was doen with the 2.4c CPU @ default speeds also with slower memory timings. I probably would get around 18000 with the 9800pro at these speeds.

I suppose you want me to run the test @ 390/310 with the 9500pro @ default speeds:D
 
N

natiels

New Member
#38
I suppose you want me to run the test @ 390/310 with the 9500pro @ default speeds
It would be interesting to see but i am guessing it would be quite similar. Extra CPU/ram speed seems to really give a boost to 3dmark01 so that pretty much explains that latest score being so much higher.
 

Associates